Comment by brian-armstrong

Text to speech is still pretty distinguishable as not-human, and that seems like an easier problem (only has to work for one specific voice, not an arbitrary voice). So just on the basis alone I wonder if this isn't still a ways out